Officer Questions
Decision Workbench
This is the officer asking-questions area, but it is not a generic chat transcript. Every response is grounded in case state, evidence, memory, actions, and audit-ready follow-up.
Pipeline: classify -> retrieve -> synthesize -> recommend -> log.
Suggested officer prompts
Start with the decisions you actually need
Non-negotiable model rules
Grounded, tool-constrained answers only
Every answer must show evidence, belief, recommended action, actions already taken, and next automated follow-up.
If evidence is weak, the response explicitly separates facts, inference, uncertainty, and recommended next check.
No uncited claims. No hallucinated action history. No implied completion without a logged event.